Mediation analysis has been studied for independent data, but relatively little work has been done for correlated data observed under stepped wedge cluster randomized trials (SW-CRTs). Motivated by the need to understand the effect mechanisms in SW-CRTs, we develop a set of regression-based methods for conducting mediation analysis. Based on linear and generalized linear mixed models, we explain how to estimate the natural indirect effect and mediation proportion in typical SW-CRTs with four data type combinations, including both continuous and binary mediators and outcomes. Furthermore, to address potential complexities due to exposure-time treatment effect heterogeneity, we further derive the mediation expressions in SW-CRTs when the total effect varies as a function of the exposure time. Simulations show that the proposed mediation estimators perform well across all data type combinations and treatment effect structures. Finally, we illustrate the use of the proposed approach in a real SW-CRT example. An R package mediateSWCRT has been developed to facilitate the practical implementation of the estimators.
Keywords: Jackknife variance; mediation analysis; mediation proportion; natural indirect effect; stepped wedge cluster randomized trials; time‐dependent treatment effect.
© 2025 John Wiley & Sons Ltd.